Build an assistant that can store and search what it sees, hears, and is told, all without leaving the device. In Building AI Assistants with On-Device Memory, you'll build a local memory system that recalls text, voice, and images, and teach it to recognize something new from a few photos. Built in partnership with @qdrant_engine and taught by @DylanCouzon, Developer Experience Engineer at Qdrant. This course also includes our new AI coding lab, so you can practice building with an AI coding agent from the DeepLearning.AI mobile app. Enroll for free: hubs.la/Q04y3S0F0

This week in The Batch, Andrew's Letter covers how open weights model GLM-5.3 nearly matched Claude Mythos at exploiting vulnerabilities, 12% vs. 14%. Also inside: 🧪 Xiaomi's new open weights leader 🎙️ Gemini 3.8 Live 💾 DeepSeek-V4.1-Flash 🔁 AREX agents 🔗 hubs.la/Q04z20mv0 #DeepLearningAI #AIEngineering #OpenWeights

Andrew Ng's AI Engineering Skills Map shows what to learn. AI Dev is where you hear from the people who built it in the real world. Keynotes from Andrew Ng and Yann LeCun. Engineers from Hugging Face, Google DeepMind, and BlackRock. 🗓️ Nov 30 to Dec 1 📍 New York City  🎟️ Grab early bird tickets at $599 while they last Get your ticket: hubs.la/Q04yWnyV0  #DeepLearningAI #AIEngineering #MachineLearning

The Data Engineering Professional Certificate is now available on DeepLearningAI. Across four courses, design and build the systems that generate, ingest, store, transform, and serve data, including batch and streaming pipelines on AWS and open-source tools. Built in partnership with @awscloud and taught by Joe Reis, co-author of Fundamentals of Data Engineering. Enroll for free: hubs.la/Q04yxqxW0

Everyone knows that applying rigid testing requirements to early stage AI projects causes them to stall… but some companies do it anyway. Andrew Ng explains why AI engineering tactics must adapt to the stage of the project, not just for speed, but for reliability. Read about how to calibrate your approach: 🛠️ Scaling evaluation pipelines and metrics 🛠️ Selecting software architecture for scale 🛠️ Structuring product feedback loops Read the full letter in The Batch: hubs.la/Q04ymLtP0 #AI #MachineLearning #TechNews #DeepLearningAI

Early stage AI projects don’t need rigid testing, but mature products do. Andrew Ng explains why AI engineering tactics must adapt to the project lifecycle. Also in this week's The Batch: 🛠️ Claude Opus 5.5 performance metrics 🛠️ Jev classification model goes viral 🛠️ Devin Fusion lead and sidekick models in one harness 🛠️ Message Passing for decentralized agents Read the full issue:hubs.la/Q04ymKnv0 #AIEngineering #MachineLearning #DeepLearningAI

Long contexts cause AI agents to forget early mistakes. Meta AI paired primary action agents with dedicated memory agents to fix context rot: 📝 Keeps structured notes on tools and past errors 🎯 Injects short reminders only at crucial moments 📈 Raised Claude Sonnet 4.5 benchmarks from 37.6% to 45.9% Read the full technical analysis:  hubs.la/Q04yf27q0 #DeepLearningAI #AIAgents #LLMs

Can you distill your way to a frontier model? 🧠 Anthropic's latest report highlights proxy query routing supporting massive distillation campaigns across commercial platforms. 🚨 Unauthorized proxies routed user prompts to Claude APIs 🛡️ This type of query forwarding creates severe data privacy risks Read our full breakdown: hubs.la/Q04x_m800 #DeepLearningAI #LLMs #AISecurity

10,000 AI agents spent 88 hours tackling Navier-Stokes equations in Lean. The resulting controversy teaches vital lessons to AI developers. 🧩 Agents can formalize complex mathematical proofs at massive scale 🧠 Human evaluation remains essential to interpret why proofs work 🔒 Enterprise data privacy and zero-data retention settings are mandatory Read our full technical analysis: hubs.la/Q04x_jNp0 #DeepLearningAI #AIAgents #AISecurity

Prompt injection defenses shouldn't rely solely on a model’s own training. Meta's Muse agent assumes the model will be tricked; instead, it builds security at the OS level. 🔑 Model never handles real credentials 🛡️ Tools run in isolated Linux containers 🛑 Independent gatekeeper verifies outbound calls Read the full analysis: hubs.la/Q04xZx500 #DeepLearningAI #AIAgents #AISecurity

Massive agent swarms define this week in AI. Here is what we covered in our twice-weekly shortform newsletter Data Points: 🧮 OpenAI used 10,000 agents to show that the Navier-Stokes equations break down, sparking a debate on prompt data privacy and the role of AI in mathematics. 📉 DeepSeek V4.1 Flash introduced a brand-new architecture for cheaper long-context workloads. Subscribe to Data Points to get technical insights delivered twice a week: hubs.la/Q04xTwMP0 #DeepLearningAI #MachineLearning #AIAgents

In this week’s letter, Andrew Ng addresses the calls from AI companies and recently departed researchers for a slowdown on development. Recent reports highlighted a swarm of 1,200 OpenAI agents compromising Hugging Face’s system. But the actual breach stemmed from inadequate sandboxing and monitoring processes. Companies need to stop assigning responsibility to runaway AI agents when it’s poor human decisions that lead to big mistakes. Read Andrew’s full argument against AI doomsayers in The Batch. hubs.la/Q04xTv-20

🛑 The latest AI doom hype is overblown. In The Batch, Andrew Ng breaks down why sloppy sandboxing, not runaway software, is the true root of recent cybersecurity incidents. Blaming too-powerful AI agents for a predictable hack is like blaming your own hammer when you foolishly break a window. Our team also breaks down the week’s most important AI news and research, including the dispute over OpenAI’s breakthrough math proof, Anthropic’s accusations against Moonshot, DeepSeek, and Alibaba, and Meta’s security protocols for its latest agent. Read the full analysis at The Batch: hubs.la/Q04xSL9J0  #DeepLearningAI #Cybersecurity #LLMs

The traditional developer role is evolving fast. With modern AI tools, the traditional lines between developer, product manager, and designer are blurring. Today's most effective AI engineers don't just implement specs, they actively shape the build. To help developers master this shift, we’ve mapped out the essential skills required to take greater ownership of the development process. Master these core skills to accelerate your workflow and deliver maximum impact: 🔄 Driving the Build Loop: Rapidly drive the build, feedback and decision loop by making informed decisions based on product vision, project stage, technical feasibility, risks, effort and budget. 🎯 Making Product Decisions: Cultivate deep user empathy and business sense to confidently steer product direction and design when a spec isn't provided. 🗣️ Communicating and leading: Act as a technical guide for your broader organization, aligning cross-functional teams (like marketing and legal) and explaining what is technically feasible. 🚀 High-Agency Ownership: Spot opportunities and deliver solutions despite ambiguity. Measure success by the value you create, not just tasks completed. Read Andrew Ng's full breakdown on shaping the build here: hubs.la/Q04xKz950  #DeepLearningAI #AIEngineering #AI

Context limits and information loss pose huge problems for long-running AI agents. 🛑 A new tool for better context management is helping developers optimize memory and cut API costs. Efficient information retrieval is a key skill for production AI. Read the full analysis in The Batch:  hubs.la/Q04xyjFN0  📖 #DeepLearningAI #AIAgents #MachineLearning

Transcription battles are heating up! 🔥 While generative text models dominate the conversation, speech recognition is having its own moment. Google, Meta, and Microsoft are actively fighting for the top spot with powerful new model releases. Learn what this means for your AI builds in The Batch: hubs.la/Q04xyhnw0  📖 #DeepLearningAI #SpeechRecognition #MachineLearning

Claude Fable 5.1 held on to the number one spot on Artificial Analysis’ Intelligence Index v4.2, and tied with OpenAI’s new model on the updated benchmarks 🏆. What you need to know: 📊 Scores 57 on the v4.2 index, edging out GPT 6 Astra. 🔬 Achieves 52.6% on Terminal Bench Science 0.1 for agentic research. 💰 Saves costs for repeated agentic workflows via cheaper cache reads. Get the full breakdown in The Batch: hubs.la/Q04xy0h-0  🔗 #DeepLearningAI #Anthropic #AIAgents

GPT-6 Astra tops the ARC-AGI-3 leaderboard while cutting token costs. Here is what matters for developers: 🚀 ⚡ Tied with Claude Fable 5.1 on Artificial Analysis’ Intelligence Index ⚡ Asynchronous tool calls for parallel execution ⚡ Retained reasoning memory across API calls Efficient context management is essential to scaling agents. Read our full analysis in The Batch: hubs.la/Q04xp3NY0  📖 #DeepLearningAI #AIAgents #MachineLearning

In his latest letter in The Batch, Andrew Ng explains why the best AI engineers do not just write code to spec. They shape the build. 🧵 Four core skills Andrew highlights to level up your engineering workflow: 🔄 Drive the build loop | Prototype fast and iterate on real user feedback. 💡 Make product decisions | Pair technical feasibility with business sense and user empathy. 📢 Communicate broadly | Align product goals across marketing, legal, and finance. ⚡ High agency ownership | Spot problems and execute solutions without waiting for top-down direction. Read Andrew's full perspective in The Batch: hubs.la/Q04xg-090 ⚡ #DeepLearningAI #AIEngineering #TechLeadership

The skills required to be an engineer in the age of AI are changing fast. Here is your quick summary of this week's issue of The Batch: 🧵 1️⃣ Shaping the Build AI engineering requires more than using AI to write code. To drive impact, developers must master: • Fast build-and-feedback loops • Product and design judgment • Stakeholder communication • High agency ownership 2️⃣ GPT-6 Astra OpenAI launched GPT-6 Astra with 1M+ input tokens and 5 reasoning levels. Benchmark data on ARC AGI 3 and other tasks shows higher per token rates can yield lower total cost per task due to better efficiency. 3️⃣ Claude Fable 5.1 Anthropic updated Fable 5.1, tying for top marks on the Artificial Analysis Intelligence Index v4.3. The model loosens some restrictions on cybersecurity tasks and is less wordy. 4️⃣ Transcription Wars Google, Meta, and Microsoft released new speech to text models. Gemini 3.5 Transcribe, Muse Voice Transcribe, and MAI-Transcribe 2 all achieved word error rates below 4%. 5️⃣ SelfCompact Johns Hopkins and Apple published SelfCompact, a method that uses explicit rubrics to prune LLM context history without parameter updates or fine-tuning. 6️⃣ Events DeepLearningAI is hosting another installment of our AI Dev developer event in NYC on November 30 and December 1. Early bird tickets are available now! Subscribe to read the full technical analyses: ⚡ hubs.la/Q04xfcHh0 #DeepLearningAI #AIEngineering #LLMs